~/collections $ cat production-readiness.mdx
Production Readiness
Curated learning path • 7 posts
Ship agents you can trust. Debug failures, evaluate quality, add disclaimers, localize for global users, and govern publishing in Microsoft 365 Copilot.
Building an agent is the easy part. Making it safe, measurable, multilingual, and shippable to thousands of users is the rest of the iceberg. This collection covers the production discipline: debugging the four most common failure patterns, evaluating quality with the Agents Evaluations CLI, adding AI disclaimers, localizing across languages with tokenized manifests, validating and repairing an inherited project, and navigating admin approval and governance to get into Microsoft 365 Copilot.
- 1· 9 min
Debugging Declarative Agent Failures in Microsoft 365 Copilot
Troubleshoot the four most common declarative agent failure patterns in Microsoft 365 Copilot: from ignored APIs to auth errors and permission issues.
- 2· 10 min
Agents Without Evals Are Just Vibes: Agents Evaluations CLI
The Agents Evaluations CLI turns declarative agent quality into repeatable tests, scored reports, and PR evidence using @microsoft/m365-copilot-eval.
- 3· 7 min
Adding AI Disclaimers to Declarative Agents
Add AI disclaimers to declarative agents to build user trust and satisfy compliance requirements. Includes full manifest config and writing tips.
- 4· 8 min
Localizing Declarative Agents: Multi-Language Agents That Actually Work
Build declarative agents that work across languages with tokenized manifests, localizationInfo, and culturally adapted instructions.
- 5· 6 min
Publishing and Governing Declarative Agents
Ship declarative agents to production in Microsoft 365 Copilot: admin approval, access control, staged rollouts, and governance best practices.
- 6· 9 min
The Declarative Agent Lifecycle, End to End with WIQD
Build, Improve, Preview, Publish. WIQD runs the whole declarative agent lifecycle as one conversation, from scaffolding through evals to store submission.
- 7· 17 min
Validating an Inherited Declarative Agent with WIQD
A manifest can be valid JSON and describe a dead agent. Use WIQD static and deep validation to triage, repair, and gate an inherited declarative agent.